Qt's corpus under its grant: the count probe holds, the split qualifies, 40 partitions are admissible¶

# research(2026-10). Qt's months 2024-10 to 2025-10, fetched under the 2026-10-15 grant from the announced address, were built with build --fewer-requests (started at 13bd2a4, before the probe gate existed), then refined, deduplicated, split, frozen and verified (2026-10-16, 23:34), and backed up. The probe the 2026-10-15 entry made a precondition then ran from committed code (comment-count-probe-qt.json, from 8.47.98.37): none of the 300 zero-count rows it asked carried a comment and every control did, so the share of skipped changes with comments is bounded at 0.0099, under COUNT_PROBE_BOUND, and the build stands without a rebuild.

The frozen corpus (datasets/gerrit-granted/qt, copied to TIGRIS as corpus-v3/qt, manifest sha256 090d39a0…) holds 1,147 pilot, 7,493 train and 809 dev examples, the test window sealed; its column joins the dataset funnel (dataset-funnel.json, docs/dataset.md), whose new first step counts the 27,858 changes the build skipped because their listing records no comment.

The registered split qualifies (split-criteria-qt.json, against OpenStack's corpus v3): 37 and 36 projects, 3,798 and 3,705 training examples, the largest project at 0.498 and 0.440 of its half, and a file-suffix total variation of 0.278 against OpenStack's ceiling of 0.319. training_size.py reads Qt's own N as 2,900; every run trains at OpenStack's 1,850, the size floor admissible_partitions.py takes. 40 partition seeds are admissible of the first 81 (admissible-partitions-qt.json), so K up to 40 is available. The development pilot follows: the first 22 admissible seeds, run k at training seed k, each scored by likelihood_score.py after its decoder check, as OpenStack's and Wikimedia's were. Qt is permitted, not admitted: its cells' bounds are simulated and committed before submission, as the gate requires.